0.5.X user DistEmbedding occur Error

Questions
,
#1

when I run unsupervised dist demo and I modify load_subtensor method about at line 245 , there occur error about process hanging.

I modify
batch_inputs = g.ndata[‘features’][input_nodes].to(device)

to
batch_inputs = emb(input_nodes).to(device)

there emb is a DistEmbedding

but why hanging 30min and occur timeout.

Is my input_nodes param need to mapping or any other thing I use wrong ?

#2

can dist embedding create on gpu direct ?

#3

another question is when I use init_function ,there occur error on graph server , and say can’t find init_emb

#4

You cannot create embedding on gpu directly.

#5

Can you provide the completed error information? Thanks!

#6

The hanging extremly likely to be caused by the failure of the server process. The input_nodes should be the local id instead of the original global id.

#7

thx for reply you are right .